Voice Spoofing in the Era of Deepfakes: Machine Learning Challenges and Solutions
Rekha Rani, Bal Kishan, Rahul · 2024
Voice spoofing has become a serious security concern due to advancements in voice cloning technologies. This review paper examines recent advances in machine learning, such as deep learning, adversarial approaches, and self-supervised learning, to identify spoof voices. We address problems including real-time detection in low-resource environments, biases, and model generalization. Our experiment on the ASV spoof 2019 dataset shows that deep learning models-particularly LSTMs-are superior to traditional models like SVMs and GMMs regarding spoof voice recognition accuracy. Conventional models are less accurate yet require more time to train. This highlights the compromise between performance and training time. We compare leading detection models and outline a plan for creating effective, fair, and scalable spoof detection systems.